umap import failure

[camneese]

I was trying to import umap into jupyter notebooks, but kept getting an error message saying numba couldn't be imported as part of the process.

The original error was: 'DLL load failed: The specified module could not be found.'

If possible please include the following in your error report:

sys.executable: C:\CodingLanguages\Anaconda\python.exe

[sklam]

If you are using numba=0.49. It could be related to #5600. Please try to use the 0.49.1rc1 wheel.

[camneese]

Thank you, that seemed to clear up the big issue! It's still throwing me one error, is this package not included in this wheel? Any way to get around it?

ModuleNotFoundError Traceback (most recent call last)
in
----> 1 import umap

~\AppData\Roaming\Python\Python37\site-packages\umap_init_.py in
----> 1 from .umap_ import UMAP
2
3 # Workaround: #3341
4 import numba
5

~\AppData\Roaming\Python\Python37\site-packages\umap\umap_.py in
24 import scipy.sparse
25 import scipy.sparse.csgraph
---> 26 import numba
27
28 import umap.distances as dist

C:\CodingLanguages\Anaconda\lib\site-packages\numba_init_.py in
18
19 # Re-export typeof
---> 20 from numba.misc.special import (
21 typeof, prange, pndindex, gdb, gdb_breakpoint, gdb_init,
22 literally, literal_unroll

ModuleNotFoundError: No module named 'numba.misc.special'

[stuartarchibald]

@camneese which version of Numba do you have? It looks like the environment might be a bit confused?

[stuartarchibald]

This is strangely similar to #5675

[stuartarchibald]

Just pip installed umap-learn==0.4.2 on a windows machine with Python 3.7, cannot reproduce the above.

[stuartarchibald]

@camneese do you have a method of reproducing this please?

[camneese]

Sorry, just saw this! I tried again a few days later and it went through without an error. I didn't do anything different; I'm still using numba=0.49.1rc1. Still not sure what caused it, but it very well may have just been a confused environment. Thanks for the help!

[esc]

@camneese cool! Thanks for reporting back and thanks for using Numba! 🙏